ShadowBench

Submit code

Generated Lean 4 solutions, evaluated for compilation and semantic alignment.

We'll email your result when evaluation finishes, with a link to publish it to the public leaderboard. Never shown or shared otherwise.
JSON or JSONL with a task_id and lean_code for each solution. Maximum 4 MiB.
No solutions loaded.
Use a password you do not use elsewhere. Deletion is checked by this site only.

Each run evaluates the same 178 tasks as the paper's test set. Missing solutions score zero.

Your submissions

Recent submissions from this browser.

No submissions yet.